Link building for the AI era

For twenty years link building followed one law: acquire more referring domains, earn more authority, rank higher. That law still holds for classic search. But a second engine now reads the web, and it does not count links the way Google does. When ChatGPT, Perplexity, Claude, or Google AI Mode assembles an answer, it leans on a small, repeated cast of trusted sources and mostly ignores the backlink long tail your team spent years building. This guide is about the new second game: earning the corroboration that gets your brand named inside AI answers, and how to run it alongside the link building that still moves rankings.

Why AI engines trust a narrow set of sources

Google's link graph is democratic in principle. Every referring domain casts a weighted vote, and a site with ten thousand modest links can outrank one with a hundred strong ones. AI engines behave differently. When a model generates an answer it retrieves a handful of documents, reads them, and cites the few it can quote cleanly. That retrieval step is not a vote count. It favors sources the model has learned are consistently reliable across its training and its live index, which collapses the web into a short list of names the engine reaches for again and again.

The practical consequence is brutal for anyone measuring success in raw link volume. A backlink from a low-authority blog might still nudge a ranking by a fraction, but it will almost never cause an engine to cite you. The engine is not asking who has the most links. It is asking which named entities are corroborated by the sources I already trust. That is why a single mention on Wikipedia, a strong G2 review profile, or a quoted stat in a trade publication can outperform hundreds of ordinary backlinks for AI visibility, even though the same asset would barely register in a traditional link audit.

There is also a structural reason. Models are tuned to avoid hallucination, so they weight sources that reduce risk. Reference sites, review platforms, established news, and primary research all carry low reputational risk for the engine. An unknown domain, however well linked, carries high risk. So the engine defaults to the safe cast. Understanding this reframes the entire discipline: for GEO you are not building a link profile, you are building a corroboration profile across the specific properties engines have decided are safe to quote.

“Google counts your links. An AI engine checks who agrees with you. Those are different games, and you have to win both.”

The six source types that earn citations

Across the AI answers we track for enterprise clients, citations cluster into six source types that do almost all the work. Community discussion, led by Reddit, gives engines the unfiltered practitioner view. Reference sites, led by Wikipedia, supply the neutral entity definition. Review platforms like G2, Capterra, and TrustRadius provide structured, comparative buyer signal. Authoritative news and trade press supply timeliness and editorial credibility. Industry associations lend institutional legitimacy. And primary research, your own or others', supplies the quotable numbers models love. The long tail of guest posts, directories, and niche blogs barely appears.

The order matters because it tells you where to spend. Community and reference together account for roughly half of citations, yet most enterprise link programs ignore both. Reddit cannot be bought and Wikipedia cannot be gamed, which is exactly why engines trust them. Reviews come third and are the most controllable of the high-value sources: you can actively drive verified reviews on G2 in a quarter. News and research require real work, a genuine story or a genuine dataset, but they compound because a single strong placement gets syndicated and re-cited for months. Note what is missing from the chart entirely. Guest posts on mid-tier blogs, paid directory listings, and the vast middle of the link graph do not appear, because engines almost never retrieve from them. If your current program spends most of its budget there, you are optimizing for a scoreboard AI engines do not read.

Digital PR and the data-led story

If community and reference sources are earned by being genuinely useful over time, the fastest lever you actively control is digital PR built on original data. The mechanism is simple. Journalists need numbers, engines love numbers, and a proprietary dataset gives you both at once. When you publish a study with a clear, surprising, defensible finding, trade press covers it, that coverage lands on authoritative news domains, and the finding itself becomes a quotable unit that AI engines lift directly into answers. One study can seed dozens of citations across the two most editorially trusted source types on the list.

The discipline is in the story, not the volume. A weak data story is a survey of two hundred people with a finding nobody disputes. A strong one pairs a real dataset, ideally something only you can measure from your product or client base, with a single headline claim a growth leader would repeat in a meeting. We look for a number that reframes a debate: a benchmark, a year-over-year shift, a counterintuitive correlation. Then we build the asset so both a journalist and a model can extract it. That means a plain headline stat, a clean methodology note, a chart with a caption, and a citation line the outlet can paste without editing.

Timing and packaging separate the placements that get cited from the ones that get buried. Give reporters an embargoed brief, a spokesperson, and the raw chart. Give engines a canonical page on your own domain with schema, a downloadable dataset, and an unambiguous cite-this line. When the same finding appears on your site, in three trade outlets, and in a syndicated newswire piece within one week, the engine sees corroboration from multiple trusted sources pointing at one claim. That convergence is what promotes you from a page that ranks to a source that gets named.

Earning corroboration versus buying links

The AI era makes the case against bought links stronger than any Google penalty ever did. Paid placements, link exchanges, and private blog networks live on exactly the low-trust domains engines have learned to discount. You can buy a hundred such links and move your citation rate by nothing, because the model never reaches for those domains when it retrieves. Worse, the entity-level trust engines assign is sticky. A brand associated with spammy corroboration gets quietly demoted as a safe source, and that reputation is far harder to repair than a ranking dip. White hat is no longer a compliance posture. It is the only approach that touches the sources that matter.

Read the table by its last two columns and the strategy writes itself. The sources with the highest AI citation weight, reference, community, and reviews, are precisely the ones you cannot buy. The tactics you can buy sit at the bottom with little ranking effect and no citation value. Earning corroboration is slower and harder than placing links, which is exactly why it works: the difficulty is the moat. When a competitor cannot simply purchase their way onto Wikipedia or into a trusted subreddit, the brand that did the real work of being useful, reviewed, and independently covered owns a position that is expensive to contest.

This does not mean abandoning legitimate outreach. Real editorial placements on real outlets, genuine partnerships, and earned mentions from people who actually use your product are corroboration, and they are white hat. The line is not paid versus unpaid, it is earned versus manufactured. A sponsored study covered on its merits is fine. A link inserted into a stranger's post for fifty dollars is not. Engines are increasingly good at telling the difference, and they resolve ambiguity in favor of the safe cast of sources they already trust.

Links that rank versus sources that get cited

The most useful mental model here is to stop treating these as one program. A link that ranks is a signal to Google's algorithm about your page's authority and relevance. A source that gets cited is a signal to a language model about which entities are safe to name. They overlap, a strong news placement can do both, but they are optimized for different scoreboards and they fail in different ways. A page can rank first in Google and never be cited by ChatGPT, because ranking rewards the link graph while citation rewards corroboration and extractability. The inverse is also true: a brand quoted constantly in AI answers can sit on page two of the SERP.

The operational takeaway is to budget against both boards on purpose. If your category's buyers increasingly open with an AI engine, the citation board deserves real investment even where it does nothing for rankings. If you still win most demand through the classic SERP, protect the link program that defends those positions. The mistake we see most often is a team running a single link retainer against a single volume target, then wondering why their AI visibility flatlines. You cannot backlink your way into an answer. You corroborate your way in.

Diagnosing where you are and are not cited

Before you earn a single new source, find out where you already stand, because the gaps tell you exactly what to build. Start with the prompts, not the keywords. List the ten to twenty questions your buying committee actually asks an AI engine across the funnel: the category definition, the best-tools-for question, the head-to-head against your main rival, the how-do-I problem query, and the is-it-worth-it evaluation. These are the moments where being cited or absent decides whether you enter the buyer's consideration set at all.

Then run each prompt across ChatGPT, Perplexity, Claude, and Google AI Mode, more than once, and log three things for every answer: are you cited at all, where do you rank in the source list, and which source is the engine actually pulling from. That last column is the diagnostic gold. If the engine cites a competitor via their G2 profile and a Reddit thread, you now know the specific corroboration gap to close. If it cites your rival's original study, you know you need a study of your own. Absence is never random; it always maps to a missing source.

Cross-reference that prompt audit with a source audit. Do you have a Wikipedia entry, and is it well sourced? What is your G2 review count and category rank against the brands that are winning the answers? When did a real journalist last quote you? Are you present, credibly, in the two or three communities where your buyers argue about tools? Score each of the six source types red, amber, or green for your brand. The reds that sit behind your highest-value prompts are your roadmap, in priority order, and they will almost always be the sources you cannot buy. A useful discipline here is to run the audit quarterly and keep the log, because citation is not a one-time state. Engines re-index, competitors earn new corroboration, and a green source can quietly slip to amber when a rival ships a better study or accumulates more recent reviews. The teams that stay cited treat the diagnosis as a standing report, not a project.

Outreach that works for B2B

Outreach for the AI era is narrower and warmer than the spray-and-pray campaigns that defined old link building. You are not chasing a hundred domains for volume, you are earning presence on a short list of specific, trusted sources tied to your priority prompts. That changes the shape of the work. Instead of a mail merge to every blog in your niche, you build relationships with the handful of reporters who cover your category, the analysts behind the review platforms, and the practitioners who moderate the communities your buyers trust. Fewer targets, higher effort each, far better return.

The angle that lands with B2B gatekeepers is always value first, and for this era that value is usually data or genuine expertise. Reporters ignore product pitches and open emails that offer an exclusive number, a fresh benchmark, or a named expert available on deadline. Community goodwill is earned by answering hard questions in threads for weeks before you ever mention your product, and by being the account people already recognize as helpful. Reviews are earned by asking happy customers at the right moment, at renewal, after a win, inside the product, not by incentive schemes that platforms flag and engines discount.

Measuring DR, referring domains, and mention rate

You still measure the classic link metrics, because they still govern rankings, but you now measure a second set alongside them. Domain Rating and referring domains remain the right lens on your link program: track them monthly, watch the trend and the quality mix, and make sure new referring domains skew toward the credible end rather than padding the count with weak sites. These numbers defend your SERP positions and they are the leading indicator that your earned-media and PR work is landing on domains worth having. Do not abandon them; just stop treating them as the whole picture.

The metric that captures the new game is mention rate: across your set of priority prompts, how often does an engine cite you at all, and where do you sit in the source list when it does. Track it weekly per engine, because the engines disagree and a gain on Perplexity does not imply a gain on ChatGPT. Pair mention rate with a source-coverage score, the red-to-green state of your six source types, so you can attribute movement to specific corroboration you earned. When a new study lands and mention rate climbs on the three prompts it addressed, you have proven the mechanism, not just watched a vanity number rise.

Report both scoreboards against one revenue number or the executive conversation falls apart. Wire DR and referring domains, mention rate and citation rank, and organic and AI-cited traffic into a single view that ties back to pipeline. The link metrics tell you whether you are defending rankings. The citation metrics tell you whether you are present in the answers that increasingly precede a ranking click. Leadership does not care which board a win came from; they care that the earned-media and corroboration budget produced pipeline. Show both moving toward one number and the program funds itself.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.